Capitalizing the Predictive Potential of Machine Learning to Detect Various Stratigraphic Units of Lunar Craters Using Multi-Source Lunar Exploration Data

This study utilized multi-class lunar exploration data including imagery and topographic data from Chang'e-2, USGS Geological Map, LROC WAC, and Kaguya MI to classify different geological unit types of lunar craters through machine learning. Focusing on the landing area of Chang'e-5 as the study region, various features such as morphology and mineral composition of craters were extracted to establish a predictive model using five machine learning algorithms including XGBoost. Results indicate that XGBoost exhibited the highest overall predictive accuracy with a classification accuracy of 97.7%, and a weighted F1 score of 97.6%. This study confirms the effectiveness of feature selection in enhancing model classification accuracy, revealing significant correlations between elevation, FeO content features, and the geological units where craters are located within the study area. The research achieved precise prediction of geological unit types within lunar craters. The predictive outcomes not only estimate the geological age represented by craters but also provide insights and references for future research on establishing quantitative analysis models for crater stratigraphic age based on morphological parameters and rock abundance.